WEBVTT 00:00:00.000 --> 00:00:03.339 I will admit something right now. I still stare 00:00:03.339 --> 00:00:05.599 at a blank page and just freeze up sometimes. 00:00:06.219 --> 00:00:09.060 It happens to all of us. Oh, absolutely. Yeah, 00:00:09.119 --> 00:00:10.500 it really does. I just sit there looking at the 00:00:10.500 --> 00:00:13.320 cursor and the cursor just blinks back at you. 00:00:13.480 --> 00:00:16.339 It expects absolute brilliance on demand. That 00:00:16.339 --> 00:00:19.539 blank page is deeply intimidating. It is. But 00:00:19.539 --> 00:00:22.879 imagine if like 95 % of your marketing work happened 00:00:22.879 --> 00:00:26.519 in just one tool. I mean, research, writing, 00:00:26.719 --> 00:00:29.480 designing, publishing. Right. Not just some basic 00:00:29.480 --> 00:00:31.679 chat bot you argue with. We are talking about 00:00:31.679 --> 00:00:34.380 a full automated team. That is exactly what we 00:00:34.380 --> 00:00:36.600 are building today. Welcome to today's deep dive. 00:00:36.759 --> 00:00:40.079 We are exploring a really fascinating architectural 00:00:40.079 --> 00:00:42.659 system. We are going to build an AI marketing 00:00:42.659 --> 00:00:45.750 team. And we are doing this inside a workspace 00:00:45.750 --> 00:00:48.250 called Codex. Yeah, and whether you are a creator, 00:00:48.409 --> 00:00:50.710 a founder, or just someone trying to scale your 00:00:50.710 --> 00:00:53.289 ideas, we want to show you a powerful workflow. 00:00:53.630 --> 00:00:55.689 You are going to move from that terrifying blank 00:00:55.689 --> 00:00:59.670 page to a massive automated content system. We 00:00:59.670 --> 00:01:01.090 are not just going to list a bunch of software 00:01:01.090 --> 00:01:03.509 either. No, definitely not. We are going to explore 00:01:03.509 --> 00:01:06.870 the actual underlying mechanics, like how do 00:01:06.870 --> 00:01:09.590 we ground the AI properly? How do we build specialized 00:01:09.590 --> 00:01:12.349 departments for research, design, and video? 00:01:12.750 --> 00:01:16.709 And how do we finally tame your chaotic inbox? 00:01:17.230 --> 00:01:19.739 Let's start with the foundation. What exactly 00:01:19.739 --> 00:01:22.879 is Codex? Because it feels fundamentally different 00:01:22.879 --> 00:01:25.200 from standard chatbots. Oh, it is completely 00:01:25.200 --> 00:01:27.500 different. Codex is not a chatbot where you just 00:01:27.500 --> 00:01:31.099 type a prompt and get text back. Codex is a super 00:01:31.099 --> 00:01:34.780 app workspace. It actually executes your tasks 00:01:34.780 --> 00:01:38.099 natively. You can preview apps right in the interface. 00:01:38.459 --> 00:01:41.480 You can edit spreadsheets dynamically. You basically 00:01:41.480 --> 00:01:44.760 call upon specific skills and plugins. Wait, 00:01:44.780 --> 00:01:46.620 let me pause right there. So skills are basically 00:01:46.620 --> 00:01:48.480 reusable instruction files, right? They tell 00:01:48.480 --> 00:01:51.060 your AI exactly what to do. Exactly. And plugins 00:01:51.060 --> 00:01:53.480 are bundles of specific skills. They are grouped 00:01:53.480 --> 00:01:55.760 together for your AI agent to use autonomously. 00:01:55.980 --> 00:01:58.340 You have a left panel for your chats. You have 00:01:58.340 --> 00:02:00.900 a middle panel for the agent's actual work. And 00:02:00.900 --> 00:02:03.459 you have a right panel for live visual preview. 00:02:03.560 --> 00:02:05.620 So Codex actually does the work instead of just 00:02:05.620 --> 00:02:08.219 talking about it. Exactly. It is a workspace 00:02:08.219 --> 00:02:12.020 where your AI actually executes the tasks. Okay. 00:02:12.060 --> 00:02:14.960 That makes sense. But to build this team, we 00:02:14.960 --> 00:02:17.500 first have to fix a glaring problem with AI. 00:02:18.240 --> 00:02:20.620 Generative models often give us incredibly generic 00:02:20.620 --> 00:02:23.219 outputs. Oh, completely. You ask for a marketing 00:02:23.219 --> 00:02:25.620 hook and it sounds completely robotic. It feels 00:02:25.620 --> 00:02:28.319 totally devoid of any real human soul. Yeah, 00:02:28.419 --> 00:02:30.680 and the fix for that is a concept called grounding. 00:02:31.280 --> 00:02:34.319 Grounding means giving your AI specific, high 00:02:34.319 --> 00:02:37.280 -quality examples to learn from first. Think 00:02:37.280 --> 00:02:39.479 about how a large language model actually works 00:02:39.479 --> 00:02:42.419 under the hood. Right. It predicts the next most 00:02:42.419 --> 00:02:45.129 likely word based on its training data. Exactly. 00:02:45.370 --> 00:02:48.090 So if you ask a standard AI for a viral intro, 00:02:48.270 --> 00:02:50.509 it just guesses based on everything. It pulls 00:02:50.509 --> 00:02:52.770 from incredibly broad Internet data. It mixes 00:02:52.770 --> 00:02:55.550 the styles of a fitness creator and a B2B sauce 00:02:55.550 --> 00:02:57.330 founder. And then it throws in a finance educator 00:02:57.330 --> 00:02:59.789 for good measure. Right. So you get one bland, 00:03:00.050 --> 00:03:02.610 completely useless output. It is kind of like 00:03:02.610 --> 00:03:05.289 handing a chef a very specific family recipe 00:03:05.289 --> 00:03:07.509 rather than just saying, make me some dinner. 00:03:07.590 --> 00:03:09.590 If you do not ground the AI, it is cooking with 00:03:09.590 --> 00:03:11.629 every single ingredient on the Internet. That 00:03:11.629 --> 00:03:14.219 is a perfect analogy. That is why you're. output 00:03:14.219 --> 00:03:17.199 tastes like a strange mix of sauce jargon and 00:03:17.199 --> 00:03:19.639 fitness tropes you absolutely have to limit the 00:03:19.639 --> 00:03:22.180 ingredients instead of letting it guess You feed 00:03:22.180 --> 00:03:24.919 the AI actual YouTube transcripts. You give it 00:03:24.919 --> 00:03:27.340 your personal swipe files. You share your specific, 00:03:27.460 --> 00:03:30.419 highly detailed brand guidelines. You give it 00:03:30.419 --> 00:03:32.860 top performing hooks that you admire. Once the 00:03:32.860 --> 00:03:35.699 AI has those specific references, it narrows 00:03:35.699 --> 00:03:38.419 its mathematical vector space. The probability 00:03:38.419 --> 00:03:41.960 of it generating your specific tone just skyrockets. 00:03:42.280 --> 00:03:44.319 So a normal workflow is just a prompt leading 00:03:44.319 --> 00:03:47.180 to a generic output. A grounded workflow finds 00:03:47.180 --> 00:03:49.539 specific structural patterns in your curated 00:03:49.539 --> 00:03:52.750 references. Yeah. That extra context changes 00:03:52.750 --> 00:03:55.289 the output quality immensely. Does grounding 00:03:55.289 --> 00:03:57.750 basically give the AI a creative compass? Right. 00:03:58.169 --> 00:04:01.289 Specific references prevent generic robotic sounding 00:04:01.289 --> 00:04:03.750 outputs completely. So since grounding requires 00:04:03.750 --> 00:04:06.710 great references, where do we get them? We build 00:04:06.710 --> 00:04:08.849 a research department. We start with the YouTube 00:04:08.849 --> 00:04:11.569 Researcher skill. It pulls video transcripts 00:04:11.569 --> 00:04:14.189 to study hooks, pacing, and specific word choice. 00:04:14.469 --> 00:04:16.649 It looks at how a successful creator transitions 00:04:16.649 --> 00:04:19.629 seamlessly between ideas. Let's look at the actual 00:04:19.629 --> 00:04:22.850 mechanism. To set this up, you use an API tool 00:04:22.850 --> 00:04:25.550 like SuperData, right? An API is basically a 00:04:25.550 --> 00:04:28.930 software bridge. Yeah, exactly. It lets two applications 00:04:28.930 --> 00:04:31.410 trade data quietly in the background. It sends 00:04:31.410 --> 00:04:34.069 a request to YouTube's server. It grabs the raw 00:04:34.069 --> 00:04:36.850 text file of the captions. It strips out all 00:04:36.850 --> 00:04:39.550 the messy timestamps. Then it feeds a clean text 00:04:39.550 --> 00:04:42.329 block right into the AI's context window. You 00:04:42.329 --> 00:04:44.290 got it. So you could study someone like Cleo 00:04:44.290 --> 00:04:47.110 Abram, and you feed her transcripts in, and the 00:04:47.110 --> 00:04:49.930 AI generates short -form hooks matching her exact 00:04:49.930 --> 00:04:52.550 rapid -fire pacing. Or let's say you pull Andres 00:04:52.550 --> 00:04:55.730 Karpathy's LLM video. Karpathy is brilliant at 00:04:55.730 --> 00:04:57.689 taking incredibly dense nodes of information 00:04:57.689 --> 00:05:00.009 and breaking them down. Oh, he really is. You 00:05:00.009 --> 00:05:02.149 learn to explain complex topics in his beginner 00:05:02.149 --> 00:05:04.350 -friendly teaching style. Because marketing is 00:05:04.350 --> 00:05:07.129 not just about writing faster. It is about understanding 00:05:07.129 --> 00:05:09.769 and transferring ideas clearly. Exactly. And 00:05:09.769 --> 00:05:12.089 once you have that external data, you need your 00:05:12.089 --> 00:05:14.449 internal data. That is where you add the ReadWise 00:05:14.449 --> 00:05:17.709 CLI skill. CLI stands for Command Line Interface. 00:05:17.889 --> 00:05:19.790 Meaning your AI can use direct text commands 00:05:19.790 --> 00:05:23.029 to control ReadWise. It skips clunky visual menus 00:05:23.029 --> 00:05:25.779 entirely. Right. And Readwise is essentially 00:05:25.779 --> 00:05:28.500 a digital second brain. It stores your saved 00:05:28.500 --> 00:05:31.220 notes, book highlights, and podcast transcripts. 00:05:31.220 --> 00:05:33.939 It holds your saved tweets and those random midnight 00:05:33.939 --> 00:05:36.699 thoughts. Because most good content does not 00:05:36.699 --> 00:05:39.420 actually start from a blank page. It starts from 00:05:39.420 --> 00:05:41.699 a fragment of something you already saved. I 00:05:41.699 --> 00:05:43.720 have to push back here, though. My Readwise is 00:05:43.720 --> 00:05:46.439 an absolute mess. I mean, whose isn't? Right. 00:05:46.480 --> 00:05:49.600 So how does the AI know which of my random half 00:05:49.600 --> 00:05:51.560 -baked saved notes are actually worth turning 00:05:51.560 --> 00:05:53.569 into content? A lot of those notes... is complete 00:05:53.569 --> 00:05:55.930 garbage right and that is exactly where skill 00:05:55.930 --> 00:05:58.129 stacking comes in skill stacking means combining 00:05:58.129 --> 00:06:00.990 multiple ai abilities into one automated workflow 00:06:00.990 --> 00:06:03.550 you combine the readwise skill with the youtube 00:06:03.550 --> 00:06:06.629 researcher skill Codex Cross references them. 00:06:06.790 --> 00:06:10.110 It finds the exact mathematical overlap. It merges 00:06:10.110 --> 00:06:12.750 your private, messy idea bank with proven market 00:06:12.750 --> 00:06:15.649 data. Oh, I see. Yeah. You ask it to review your 00:06:15.649 --> 00:06:17.790 saved items from the past week. Then you ask 00:06:17.790 --> 00:06:20.189 it to study the transcripts of your last 10 successful 00:06:20.189 --> 00:06:23.089 videos. The AI finds the semantic overlap. Then 00:06:23.089 --> 00:06:25.250 it generates 20 new video ideas that are both 00:06:25.250 --> 00:06:27.769 deeply personal and statistically proven. It 00:06:27.769 --> 00:06:30.829 merges my personal taste with proven market data. 00:06:31.069 --> 00:06:34.290 Exactly. Your personal ideas combined with proof. 00:06:34.410 --> 00:06:38.910 Okay, so now we have researched solid data -backed 00:06:38.910 --> 00:06:41.730 ideas, but we need to communicate them. Plain 00:06:41.730 --> 00:06:43.850 text is not always enough for an audience. We 00:06:43.850 --> 00:06:45.930 need to build a design department. For that, 00:06:46.009 --> 00:06:48.990 we use the ExcalDraw diagram skill. ExcalDraw 00:06:48.990 --> 00:06:51.449 creates these simple, unpolished visual structures. 00:06:51.550 --> 00:06:54.290 It uses boxes, simple arrows, and minimal text. 00:06:54.649 --> 00:06:56.490 Under the hood, ExcalDraw is really just generating 00:06:56.490 --> 00:06:59.699 a JSON file, isn't it? Yeah, exactly. The AI 00:06:59.699 --> 00:07:02.819 writes code dictating coordinate points. Codex 00:07:02.819 --> 00:07:04.779 then renders those coordinates as a visual diagram 00:07:04.779 --> 00:07:08.360 on your screen. It is phenomenal for explaining 00:07:08.360 --> 00:07:11.019 abstract concepts very quickly. For example, 00:07:11.100 --> 00:07:13.060 you can visually explain how your skills and 00:07:13.060 --> 00:07:15.220 plugins connect. Right. You just need the clear, 00:07:15.259 --> 00:07:18.139 bare -bones shape of the idea. So Excalibur is 00:07:18.139 --> 00:07:20.139 like the quick whiteboard sketch you do with 00:07:20.139 --> 00:07:22.160 a colleague over coffee. It's about clarity. 00:07:22.540 --> 00:07:25.819 But sometimes you need real polish. We all know 00:07:25.819 --> 00:07:28.019 the danger of overdesigning a simple concept. 00:07:28.240 --> 00:07:30.560 I have definitely spent three hours tweaking 00:07:30.560 --> 00:07:32.319 a graphic that should have been a napkin sketch. 00:07:32.600 --> 00:07:35.360 We have all been there. But when you do need 00:07:35.360 --> 00:07:38.180 that final polish, that brings us to the paper 00:07:38.180 --> 00:07:42.459 MCP skill. MCP stands for Model Context Protocol. 00:07:43.050 --> 00:07:45.430 It is basically a standardized protocol letting 00:07:45.430 --> 00:07:48.470 AI safely operate external software tools. Right. 00:07:48.569 --> 00:07:51.670 And PaperMCP is for polished UI and high -end 00:07:51.670 --> 00:07:54.769 layouts. Think animated explainers and gorgeous 00:07:54.769 --> 00:07:57.589 website mock -ups. It is perfect for YouTube 00:07:57.589 --> 00:08:00.610 thumbnail concepts and slick Instagram carousels. 00:08:01.370 --> 00:08:03.810 Excalidraw is for quick, structural thinking. 00:08:04.129 --> 00:08:06.370 Paper is for when the visual needs to be shared 00:08:06.370 --> 00:08:08.629 publicly with clients. And you can use subagents 00:08:08.629 --> 00:08:11.089 to speed this entire process up, right? Subagents 00:08:11.089 --> 00:08:13.480 are smaller, specialized. AI helpers handling 00:08:13.480 --> 00:08:16.480 parallel tasks. Exactly. One subagent is researching 00:08:16.480 --> 00:08:18.279 the color palette while another is building the 00:08:18.279 --> 00:08:19.959 layout structure. And with paper, you get live 00:08:19.959 --> 00:08:22.279 steering. Live steering. Yeah. You can literally 00:08:22.279 --> 00:08:24.379 watch the AI design the graphic in the right 00:08:24.379 --> 00:08:26.860 panel. You get feedback mid -process. You just 00:08:26.860 --> 00:08:29.220 type, fix the overlap on the top left, make the 00:08:29.220 --> 00:08:31.800 label slightly shorter, the AI adjusts the code, 00:08:31.860 --> 00:08:34.700 and the design updates instantly. You guide it 00:08:34.700 --> 00:08:36.639 exactly like a creative director looking over 00:08:36.639 --> 00:08:38.840 a designer's shoulder. So we use Excalibur for 00:08:38.840 --> 00:08:41.379 clarity and paper for polish. Yeah. Sketch the 00:08:41.379 --> 00:08:45.610 structure. Static visuals are incredibly useful, 00:08:45.750 --> 00:08:48.590 but modern marketing constantly demands motion. 00:08:49.000 --> 00:08:51.419 We need quick iteration and dynamic content. 00:08:51.639 --> 00:08:54.100 We need to build the video and media lab. This 00:08:54.100 --> 00:08:56.820 is where things get really advanced. We use tools 00:08:56.820 --> 00:08:59.759 called Remotion and Hyperframes. Remotion provides 00:08:59.759 --> 00:09:03.259 clean, professional UI and overlays. Think of 00:09:03.259 --> 00:09:06.159 a slick seven -section YouTube intro graphic. 00:09:06.519 --> 00:09:09.019 And Hyperframes handles advanced motion and complex, 00:09:09.059 --> 00:09:11.720 realistic physics. Right. What is fascinating 00:09:11.720 --> 00:09:14.500 is that Remotion is entirely code -based. It 00:09:14.500 --> 00:09:17.559 uses React. The AI is not dragging and dropping 00:09:17.559 --> 00:09:19.799 clips on a time machine. It is writing React 00:09:19.799 --> 00:09:22.019 components that mathematically define where a 00:09:22.019 --> 00:09:24.500 visual element lives at any given millisecond. 00:09:24.580 --> 00:09:27.700 Whoa. Beat. Imagine generating complex video 00:09:27.700 --> 00:09:30.259 keyframes. Like an animated phone flying in, 00:09:30.320 --> 00:09:31.980 showing a scrolling group chat, and zooming out 00:09:31.980 --> 00:09:34.279 into a logo. Just by typing a single descriptive 00:09:34.279 --> 00:09:37.240 sentence. Beat. It honestly feels a little bit 00:09:37.240 --> 00:09:39.889 like magic. It changes everything about media 00:09:39.889 --> 00:09:42.429 production. Because it is code, you can tell 00:09:42.429 --> 00:09:44.730 the AI to change the background gradient color 00:09:44.730 --> 00:09:47.669 exactly at the 10 -second mark. You can add a 00:09:47.669 --> 00:09:51.169 smooth ease -in spin right before the exit transition. 00:09:51.470 --> 00:09:53.309 Right. You do not have to build those keyframes 00:09:53.309 --> 00:09:56.470 manually ever again. You can also reuse old templates 00:09:56.470 --> 00:09:59.029 effortlessly. You just ask Codex to update the 00:09:59.029 --> 00:10:01.690 text and swap the brand colors. Then you export 00:10:01.690 --> 00:10:04.289 the final render straight into Premiere Pro or 00:10:04.289 --> 00:10:07.309 CapCut for the final polish. Exactly. Now let's 00:10:07.309 --> 00:10:10.070 talk about the Gen Media Mini app. A mini app 00:10:10.070 --> 00:10:12.690 in Codex is a shared visual workspace for you 00:10:12.690 --> 00:10:15.909 and the AI agent. This specific app uses the 00:10:15.909 --> 00:10:19.169 FAL API for its media generation. I want to explain 00:10:19.169 --> 00:10:21.850 the mechanism here. The FAL API is essentially 00:10:21.850 --> 00:10:24.629 a lightning -fast cloud engine. It handles the 00:10:24.629 --> 00:10:26.850 massive computational weight of rendering generative 00:10:26.850 --> 00:10:29.629 media almost instantly. Yeah, so your local machine 00:10:29.629 --> 00:10:32.190 doesn't crash. And how does that specific media 00:10:32.190 --> 00:10:35.090 workflow actually look in practice? The AI generates 00:10:35.090 --> 00:10:37.549 four different thumbnail options based on specific 00:10:37.549 --> 00:10:40.409 reference images. For example, it mimics the 00:10:40.409 --> 00:10:42.889 bold, contrasting visual style of Matt Wolfe. 00:10:43.009 --> 00:10:45.789 The options populate immediately in a visual 00:10:45.789 --> 00:10:48.889 grid within the mini app. It stores your image 00:10:48.889 --> 00:10:51.470 -to -video outputs and upscaled videos all in 00:10:51.470 --> 00:10:54.019 one place. The visual grid is a game changer. 00:10:54.159 --> 00:10:56.019 The human looks at the grid and picks the best 00:10:56.019 --> 00:10:58.899 one. Then you ask the AI to refine that specific 00:10:58.899 --> 00:11:01.600 choice. You tell it to dim the background lighting. 00:11:01.879 --> 00:11:05.200 You add bold white text to the foreground. You 00:11:05.200 --> 00:11:07.539 make the main subject pop with higher contrast. 00:11:07.960 --> 00:11:09.779 You do not have to guess what the prompt will 00:11:09.779 --> 00:11:12.679 do. You iterate visually. Exactly. Visual grid 00:11:12.679 --> 00:11:15.000 stops us from endless scrolling and chat. Yes. 00:11:15.480 --> 00:11:18.159 Side -by -side comparison speeds up your creative 00:11:18.159 --> 00:11:21.480 decisions. We'll be right back after this short 00:11:21.480 --> 00:11:26.539 break. And we are back. The content is researched, 00:11:26.779 --> 00:11:29.299 designed, and fully rendered. Now we have to 00:11:29.299 --> 00:11:31.159 manage the actual business side of things. We 00:11:31.159 --> 00:11:33.100 have to distribute the work and handle the inbound 00:11:33.100 --> 00:11:35.220 traffic. It is time for the operations department. 00:11:35.500 --> 00:11:38.159 We use the email manager or the brand deal manager 00:11:38.159 --> 00:11:41.200 skill. It actively searches your inbox for valuable 00:11:41.200 --> 00:11:44.320 sponsorships. We all know your inbox gets incredibly 00:11:44.320 --> 00:11:48.039 messy very fast. It is a chaotic stream of unstructured 00:11:48.039 --> 00:11:50.470 text. Missing the right email can mean missing 00:11:50.470 --> 00:11:53.389 a massive brand deal. Oh, absolutely. I want 00:11:53.389 --> 00:11:56.110 to focus on the logic here. It doesn't just read 00:11:56.110 --> 00:11:59.269 words. It filters out the noise by running logical 00:11:59.269 --> 00:12:02.370 operations. It removes duplicate threads. It 00:12:02.370 --> 00:12:04.509 checks the brand's audience fit against your 00:12:04.509 --> 00:12:08.110 core demographics. Yes. It actively filters for 00:12:08.110 --> 00:12:10.470 actual paid opportunities versus people just 00:12:10.470 --> 00:12:13.909 asking for free exposure. It parses all that 00:12:13.909 --> 00:12:16.409 unstructured data into a highly organized priority 00:12:16.409 --> 00:12:19.710 table. The table explicitly shows the brand, 00:12:19.870 --> 00:12:22.730 the offer amount, the audience fit, the priority 00:12:22.730 --> 00:12:25.750 level, and the exact next step. For example, 00:12:25.889 --> 00:12:28.629 it flags a high -paying sauce tool as an immediate 00:12:28.629 --> 00:12:32.049 high priority. It flags a random, vague agency 00:12:32.049 --> 00:12:34.889 email as a very low priority. It even integrates 00:12:34.889 --> 00:12:37.120 directly with your account? API to find open 00:12:37.120 --> 00:12:39.620 slots for intro calls. Yeah. It's incredibly 00:12:39.620 --> 00:12:42.360 organized. But wait, if I let an AI filter my 00:12:42.360 --> 00:12:44.620 inbox, isn't there a massive risk it archives 00:12:44.620 --> 00:12:47.320 a $10 ,000 brand deal just because the email 00:12:47.320 --> 00:12:49.659 was formatted weirdly? That is a totally valid 00:12:49.659 --> 00:12:52.259 concern, which is why the AI does not delete 00:12:52.259 --> 00:12:55.580 anything. It simply labels and categorizes. It 00:12:55.580 --> 00:12:58.019 builds the priority table for your review. You 00:12:58.019 --> 00:13:00.200 still see everything, but the high value signals 00:13:00.200 --> 00:13:01.860 are pushed to the top of the pile. That makes 00:13:01.860 --> 00:13:04.080 sense. What about publishing the actual content 00:13:04.080 --> 00:13:07.240 we created? use the Buffer publisher skill. It 00:13:07.240 --> 00:13:10.259 moves the generated approved drafts out of the 00:13:10.259 --> 00:13:12.519 Codex chat environment. It places them straight 00:13:12.519 --> 00:13:14.980 into a structured scheduling queue. Good ideas 00:13:14.980 --> 00:13:17.600 completely disappear if they stay buried in old 00:13:17.600 --> 00:13:20.120 chat threads. Buffer acts as a dedicated holding 00:13:20.120 --> 00:13:22.779 space. Right. You review your recent research. 00:13:22.899 --> 00:13:25.220 You choose the five absolute strongest ideas 00:13:25.220 --> 00:13:28.440 for LinkedIn. The AI drafts the posts and adds 00:13:28.440 --> 00:13:30.740 them securely to Buffer. I love this because 00:13:30.740 --> 00:13:33.100 it forces a separation between the creation brain 00:13:33.100 --> 00:13:35.940 and the publishing brain. Trying to do both at 00:13:35.940 --> 00:13:38.240 the exact same time is a guaranteed recipe for 00:13:38.240 --> 00:13:41.580 creative burnout. But there is a vital non -negotiable 00:13:41.580 --> 00:13:44.539 rule here. You must never automate the final 00:13:44.539 --> 00:13:47.799 send. Never. The AI drafts the sponsor reply. 00:13:48.139 --> 00:13:51.259 The AI queues the social post. But the human 00:13:51.259 --> 00:13:54.600 always clicks approve. Your money, your reputation, 00:13:54.679 --> 00:13:57.519 and your relationships are on the line. Exactly. 00:13:57.580 --> 00:14:00.440 So it acts as a highly organized gatekeeper for 00:14:00.440 --> 00:14:03.100 my inbox. Great. It sorts out the noise. You 00:14:03.100 --> 00:14:05.120 make the final call. Hearing all these different 00:14:05.120 --> 00:14:08.379 skills and departments can feel completely overwhelming. 00:14:09.220 --> 00:14:11.759 How do we actually build this system without 00:14:11.759 --> 00:14:14.200 breaking our current workflow? Well, the core 00:14:14.200 --> 00:14:16.919 building principle is patience. Do not try to 00:14:16.919 --> 00:14:19.580 build 20 skills on day one. That creates a very 00:14:19.580 --> 00:14:22.820 messy, fragile system. The progression should 00:14:22.820 --> 00:14:25.090 be simple and deliberate. Start with a single 00:14:25.090 --> 00:14:27.809 manual task. Improve the prompt carefully over 00:14:27.809 --> 00:14:30.809 a few days. Once the output is excellent, save 00:14:30.809 --> 00:14:33.509 it as a reusable skill. You just tell Codex to 00:14:33.509 --> 00:14:36.190 save that exact workflow for future use. And 00:14:36.190 --> 00:14:37.950 you only automate the process once the output 00:14:37.950 --> 00:14:40.889 is reliably good. If the output is messy, automation 00:14:40.889 --> 00:14:43.429 just creates messy work much faster. Yeah. You 00:14:43.429 --> 00:14:45.490 can automate a daily read -wise summary to hit 00:14:45.490 --> 00:14:47.889 your inbox at 8 in the morning. But you should 00:14:47.889 --> 00:14:50.269 only do that when you actually love reading the 00:14:50.269 --> 00:14:52.850 daily output. Exactly. And that brings us to 00:14:52.850 --> 00:14:55.669 the... Ultimate big idea of this deep dive, skill 00:14:55.669 --> 00:14:58.750 stacking. One single skill is definitely helpful, 00:14:58.950 --> 00:15:02.029 but chaining the YouTube researcher plus ReadWise 00:15:02.029 --> 00:15:05.330 plus Excalidraw is absolute magic. It equals 00:15:05.330 --> 00:15:08.490 a fully automated, unstoppable assembly line. 00:15:08.710 --> 00:15:10.789 We really must highlight the final 10 % rule. 00:15:11.190 --> 00:15:14.409 The AI handles the heavy lifting, the unstructured 00:15:14.409 --> 00:15:16.789 data sorting, and the busy work. It creates the 00:15:16.789 --> 00:15:19.009 rough first drafts and the various visual options. 00:15:19.070 --> 00:15:22.600 But you are the creative director. Taste. judgment, 00:15:22.639 --> 00:15:25.200 and high -level strategy strictly belong to the 00:15:25.200 --> 00:15:28.059 human. The AI simply gives you more cognitive 00:15:28.059 --> 00:15:30.759 space to think clearly. It does. So my role shifts 00:15:30.759 --> 00:15:32.460 from being the intern to the creative director. 00:15:32.659 --> 00:15:35.080 Exactly. The AI grinds through the work. You 00:15:35.080 --> 00:15:37.679 direct the vision. Let us bring this deep dive 00:15:37.679 --> 00:15:39.600 to a close. I want you to think about this for 00:15:39.600 --> 00:15:42.080 a second. If an AI system can perfectly mimic 00:15:42.080 --> 00:15:44.159 our pacing patterns, parse our private notes, 00:15:44.299 --> 00:15:47.259 and flawlessly design our graphics, does our 00:15:47.259 --> 00:15:50.759 true enduring value as creators actually come 00:15:50.759 --> 00:15:52.840 from our imperfections? Maybe the completely 00:15:52.840 --> 00:15:55.700 unexpected, unpatterned connections that only 00:15:55.700 --> 00:15:58.159 human intuition can make are the only things 00:15:58.159 --> 00:16:01.179 AI will never be able to replicate. Two sec silence. 00:16:01.559 --> 00:16:04.179 That is a profound way to look at it. The imperfections 00:16:04.179 --> 00:16:07.090 are the signature. Pick just one repetitive task 00:16:07.090 --> 00:16:09.710 today. Take something that drains your energy. 00:16:09.950 --> 00:16:12.370 Run it through an AI, refine the prompt, and 00:16:12.370 --> 00:16:14.070 see what happens. Just start building the foundation. 00:16:14.549 --> 00:16:16.409 Thanks for joining us. We will catch you on the 00:16:16.409 --> 00:16:16.970 next deep dive.